Rapid Characterization of Store Trajectory Dynamics Using Reservoir Computing

The six degrees-of-freedom (6DOF) motion of a store released from an aircraft cavity can be exceedingly complex, especially for vehicles operating at supersonic or hypersonic velocities.  In these conditions, the store must pass through shear flows, boundary layer separation regions, vortices, and turbulence, each of which can alter the trajectory in subtle and unexpected ways.  In extreme cases, the store can fail to separate or reverse course back up toward the aircraft, condition called a “bad launch."  It is of great interest to the Air Force to identify the fundamental causes of these trajectory dynamics so that rules can be established for cavity and store designs for the next generation of attack aircraft.  These new fighters, bombers, and remotely piloted aircraft (RPA) will be designed for weapons deliveries at higher altitudes and Mach numbers than their predecessors and will require extensive simulation and testing to validate stores release functionality in the expanded envelope. Current approaches to understanding store trajectory dynamics rely on live-fly observations, wind tunnel tests, and computational fluid dynamics (CFD) simulations.  CFD simulations are often the starting point of any design cycle because they do not require building a physical system and reduce the need for repeated experimental trials.  Unfortunately, state-of-the-art (SOTA) CFD simulations usually require super-computer-level resources for generating example launch events and can take weeks of computer time.  And if results are unexpected or undesired (e.g. a bad launch), the required analysis of large amounts of data may reveal no obvious root cause in the conditions that govern store trajectory.  In short, the discrepancies are so small, and the underlying physics so complex, that current data analysis methods cannot determine the signatures of a bad launch. ResCon Technologies, LLC and Clarkson University propose collaborating with the Air Force to explore a pioneering new approach: Machine Accelerated Analysis and Prediction (MAAP), a machine learning (ML)-based solution with the ability to analyze, predict, and generalize the dynamics of a system based on limited CFD results.  MAAP is built on our recent ML breakthrough known as Next-Generation Reservoir Computing (NG-RC) and will rapidly learn cavity/store dynamics to (a) provide a computationally efficient framework for evaluating the effects of various launch parameters, and (b) accelerate the solution of CFD problems.  MAAP will have the power to inform both aircraft design and weapons release control schemes while saving time, energy, and money over current methodologies.  Ultimately, MAAP’s advancements will lower risk to the warfighter, increase lethality, and lead to a greater probability of mission success.